A trained neural network watches your hands, not the art. Each webcam frame is fed to Google's MediaPipe gesture & hand model, which locates 21 3D points per hand in real time, entirely in your browser with no server.
The app reads those points: your index fingertip is the brush, and a pinch (thumb + index together) is the pen. Gestures like ✊ fist and ✋ open palm are classified by the model to clear and erase. A one-euro filter smooths the tracking so lines are clean. Symmetry mirrors every stroke into a live kaleidoscope, and Body mode swaps in a full-body pose model so you paint with your whole body.
The machine doesn't make the art; you do. But it couldn't exist without it: there's no way to paint in mid-air unless a neural network perceives your motion about 60 times a second. Built for Hack the Arts 2026.